Next month, Cisco plans to introduce an AI-powered analytics engine that predicts network problems based on abnormalities discovered in traffic patterns.
The company will unveil the engine, which has been in development for the past two years, in the form of a SaaS product at Cisco Live next month. Cisco expects to integrate the technology into a wide range of products over the next couple of years.
To develop the engine, Cisco used machine learning to identify patterns of signals that generally occur before a network path goes down or significantly degrades. When the predictive engine spots warning signs like the degradation of key performance indicators, it will advise a network manager to route traffic over a different path before the path in question degrades enough to affect the end user. The predictions range from hours to weeks before a potential incident.
Cisco has trained the predictive engine on extensive data sets. However, it still ingests telemetry from a customer's network to better tailor predictions to the network's specific behavior. Right now, the company's priority isn't for the tool to catch every potential issue but instead to build confidence by making highly accurate predictions.
"If you make a lot of false positives, then you'll lose the user's trust of the system," Cisco Engineering Fellow JP Vasseur said. "[So] when you make a prediction, you must be right."
Vasseur declined to put an exact number on the predictive engine's overall accuracy rate, as results would vary by network. However, an early trial customer had reported an accuracy rate of 99%, he said.
Networking professionals spend 41% of their time on troubleshooting and capacity management, according to a recent survey of 409 respondents by Enterprise Management Associates (EMA). Both are tasks that a predictive engine would address. End users report roughly a third of IT service problems before network managers are aware of the issue.
The same research found that 61% of networking pros rated predictive analysis as the capability that needed the most improvement. As a SaaS offering, the predictive engine should ease concerns about the burden of managing the tool, said Shamus McGillicuddy, an analyst at EMA. But putting the technology in the cloud does introduce data privacy problems for companies.
A third of enterprises would not use a tool that uploads their data for any purpose. An additional half would require strong assurances from the vendor that it would protect their data, McGillicuddy said. Cisco would need a sound cybersecurity plan and solid role-based access controls for the customized SaaS tool to succeed.
But if Cisco can tackle those potential stumbling blocks, there would be an appetite for the tool.
"The No. 1 wish from users of AI and ML [network management tools] is, we'd love to see people get better at predictive analytics with networks," McGillicuddy said. "Cisco is tackling something that's a need in the industry, that's for sure."
Madelaine Millar is a news writer covering network technology at TechTarget. She has previously written about science and technology for MIT's Lincoln Laboratory and the Khoury College of Computer Sciences, as well as covering community news for Boston Globe Media.